Within this paper we proposed a new method named BayesACO, to improve the convolutional neural network based on neural architecture search with hyperparameters optimization. At its essence BayesACO in first side uses Ant Colony Optimization (ACO) to generate the best neural architecture. In other side, it uses bayesian hyperparameters optimization to select the best hyperparameters. We applied this method on Mnist and FashionMnist datasets. Our proposed method proven competitive results with other methods of convolutional neural network optimization.